Fabrix-AI-Inc/Triton-VX-Qwen3.5-2B-DPO-LoRA

Triton VX Qwen3.5 2B (DPO) - LoRA Adapter

LoRA adapter weights for Triton VX Qwen3.5 2B, trained with Direct Preference Optimization (DPO) for Fabrix.ai / RDAF citation retrieval.

• —Base Model: unsloth/Qwen3.5-2B (or checkpoint-300 SFT model)
• —Adapter Type: PEFT LoRA (r=16, alpha=32, targetmodules: `qproj, kproj, vproj, oproj, gateproj, upproj, downproj`)
• —Full Merged Standalone Model: `Fabrix-AI-Inc/Triton-VX-Qwen3.5-2B-DPO`